assert(var->data.location >= 0);
const struct glsl_type *type = var->type;
- if (nir_is_per_vertex_io(var, stage)) {
+ if (nir_is_per_vertex_io(var, stage) || var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
*
* Example usage is:
*
- * progress = nir_remove_unused_io_vars(producer,
- * &producer->outputs,
+ * progress = nir_remove_unused_io_vars(producer, nir_var_shader_out,
* read, patches_read) ||
* progress;
*
* variable is used!
*/
bool
-nir_remove_unused_io_vars(nir_shader *shader, struct exec_list *var_list,
+nir_remove_unused_io_vars(nir_shader *shader,
+ nir_variable_mode mode,
uint64_t *used_by_other_stage,
uint64_t *used_by_other_stage_patches)
{
bool progress = false;
uint64_t *used;
- nir_foreach_variable_safe(var, var_list) {
+ assert(mode == nir_var_shader_in || mode == nir_var_shader_out);
+
+ nir_foreach_variable_with_modes_safe(var, shader, mode) {
if (var->data.patch)
used = used_by_other_stage_patches;
else
var->data.location = 0;
var->data.mode = nir_var_shader_temp;
- exec_node_remove(&var->node);
- exec_list_push_tail(&shader->globals, &var->node);
-
progress = true;
}
}
uint64_t read[4] = { 0 }, written[4] = { 0 };
uint64_t patches_read[4] = { 0 }, patches_written[4] = { 0 };
- nir_foreach_variable(var, &producer->outputs) {
+ nir_foreach_shader_out_variable(var, producer) {
for (unsigned i = 0; i < get_num_components(var); i++) {
if (var->data.patch) {
patches_written[var->data.location_frac + i] |=
}
}
- nir_foreach_variable(var, &consumer->inputs) {
+ nir_foreach_shader_in_variable(var, consumer) {
for (unsigned i = 0; i < get_num_components(var); i++) {
if (var->data.patch) {
patches_read[var->data.location_frac + i] |=
tcs_add_output_reads(producer, read, patches_read);
bool progress = false;
- progress = nir_remove_unused_io_vars(producer, &producer->outputs, read,
+ progress = nir_remove_unused_io_vars(producer, nir_var_shader_out, read,
patches_read);
- progress = nir_remove_unused_io_vars(consumer, &consumer->inputs, written,
+ progress = nir_remove_unused_io_vars(consumer, nir_var_shader_in, written,
patches_written) || progress;
return progress;
* TODO: allow better packing of complex types.
*/
static void
-get_unmoveable_components_masks(struct exec_list *var_list,
+get_unmoveable_components_masks(nir_shader *shader,
+ nir_variable_mode mode,
struct assigned_comps *comps,
gl_shader_stage stage,
bool default_to_smooth_interp)
{
- nir_foreach_variable_safe(var, var_list) {
+ nir_foreach_variable_with_modes_safe(var, shader, mode) {
assert(var->data.location >= 0);
/* Only remap things that aren't built-ins. */
var->data.location - VARYING_SLOT_VAR0 < MAX_VARYINGS_INCL_PATCH) {
const struct glsl_type *type = var->type;
- if (nir_is_per_vertex_io(var, stage)) {
+ if (nir_is_per_vertex_io(var, stage) || var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
}
static void
-remap_slots_and_components(struct exec_list *var_list, gl_shader_stage stage,
+remap_slots_and_components(nir_shader *shader, nir_variable_mode mode,
struct varying_loc (*remap)[4],
uint64_t *slots_used, uint64_t *out_slots_read,
uint32_t *p_slots_used, uint32_t *p_out_slots_read)
{
+ const gl_shader_stage stage = shader->info.stage;
uint64_t out_slots_read_tmp[2] = {0};
uint64_t slots_used_tmp[2] = {0};
/* We don't touch builtins so just copy the bitmask */
slots_used_tmp[0] = *slots_used & BITFIELD64_RANGE(0, VARYING_SLOT_VAR0);
- nir_foreach_variable(var, var_list) {
+ nir_foreach_variable_with_modes(var, shader, mode) {
assert(var->data.location >= 0);
/* Only remap things that aren't built-ins */
var->data.location - VARYING_SLOT_VAR0 < MAX_VARYINGS_INCL_PATCH) {
const struct glsl_type *type = var->type;
- if (nir_is_per_vertex_io(var, stage)) {
+ if (nir_is_per_vertex_io(var, stage) || var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
/* Count the number of varying that can be packed and create a mapping
* of those varyings to the array we will pass to qsort.
*/
- nir_foreach_variable(var, &producer->outputs) {
+ nir_foreach_shader_out_variable(var, producer) {
/* Only remap things that aren't builtins. */
if (var->data.location >= VARYING_SLOT_VAR0 &&
continue;
const struct glsl_type *type = var->type;
- if (nir_is_per_vertex_io(var, producer->info.stage)) {
+ if (nir_is_per_vertex_io(var, producer->info.stage) || var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
if (!vc_info->initialised) {
const struct glsl_type *type = in_var->type;
- if (nir_is_per_vertex_io(in_var, consumer->info.stage)) {
+ if (nir_is_per_vertex_io(in_var, consumer->info.stage) ||
+ in_var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
struct assigned_comps *assigned_comps,
bool default_to_smooth_interp)
{
- struct exec_list *input_list = &consumer->inputs;
- struct exec_list *output_list = &producer->outputs;
struct varying_loc remap[MAX_VARYINGS_INCL_PATCH][4] = {{{0}, {0}}};
struct varying_component *varying_comp_info;
unsigned varying_comp_info_size;
uint64_t zero = 0;
uint32_t zero32 = 0;
- remap_slots_and_components(input_list, consumer->info.stage, remap,
+ remap_slots_and_components(consumer, nir_var_shader_in, remap,
&consumer->info.inputs_read, &zero,
&consumer->info.patch_inputs_read, &zero32);
- remap_slots_and_components(output_list, producer->info.stage, remap,
+ remap_slots_and_components(producer, nir_var_shader_out, remap,
&producer->info.outputs_written,
&producer->info.outputs_read,
&producer->info.patch_outputs_written,
struct assigned_comps assigned_comps[MAX_VARYINGS_INCL_PATCH] = {{0}};
- get_unmoveable_components_masks(&producer->outputs, assigned_comps,
+ get_unmoveable_components_masks(producer, nir_var_shader_out,
+ assigned_comps,
producer->info.stage,
default_to_smooth_interp);
- get_unmoveable_components_masks(&consumer->inputs, assigned_comps,
+ get_unmoveable_components_masks(consumer, nir_var_shader_in,
+ assigned_comps,
consumer->info.stage,
default_to_smooth_interp);
{
nir_variable *input_vars[MAX_VARYING] = { 0 };
- nir_foreach_variable(var, &consumer->inputs) {
+ nir_foreach_shader_in_variable(var, consumer) {
if (var->data.location >= VARYING_SLOT_VAR0 &&
var->data.location - VARYING_SLOT_VAR0 < MAX_VARYING) {
}
}
- nir_foreach_variable(var, &producer->outputs) {
+ nir_foreach_shader_out_variable(var, producer) {
if (var->data.location >= VARYING_SLOT_VAR0 &&
var->data.location - VARYING_SLOT_VAR0 < MAX_VARYING) {
static nir_variable *
get_matching_input_var(nir_shader *consumer, nir_variable *out_var)
{
- nir_foreach_variable(var, &consumer->inputs) {
+ nir_foreach_shader_in_variable(var, consumer) {
if (does_varying_match(out_var, var))
return var;
}
static void
insert_sorted(struct exec_list *var_list, nir_variable *new_var)
{
- nir_foreach_variable(var, var_list) {
+ nir_foreach_variable_in_list(var, var_list) {
if (var->data.location > new_var->data.location) {
exec_node_insert_node_before(&var->node, &new_var->node);
return;
}
static void
-sort_varyings(struct exec_list *var_list)
+sort_varyings(nir_shader *shader, nir_variable_mode mode,
+ struct exec_list *sorted_list)
{
- struct exec_list new_list;
- exec_list_make_empty(&new_list);
- nir_foreach_variable_safe(var, var_list) {
+ exec_list_make_empty(sorted_list);
+ nir_foreach_variable_with_modes_safe(var, shader, mode) {
exec_node_remove(&var->node);
- insert_sorted(&new_list, var);
+ insert_sorted(sorted_list, var);
}
- exec_list_move_nodes_to(&new_list, var_list);
}
void
-nir_assign_io_var_locations(struct exec_list *var_list, unsigned *size,
- gl_shader_stage stage)
+nir_assign_io_var_locations(nir_shader *shader, nir_variable_mode mode,
+ unsigned *size, gl_shader_stage stage)
{
unsigned location = 0;
unsigned assigned_locations[VARYING_SLOT_TESS_MAX];
uint64_t processed_locs[2] = {0};
- sort_varyings(var_list);
+ struct exec_list io_vars;
+ sort_varyings(shader, mode, &io_vars);
int UNUSED last_loc = 0;
bool last_partial = false;
- nir_foreach_variable(var, var_list) {
+ nir_foreach_variable_in_list(var, &io_vars) {
const struct glsl_type *type = var->type;
- if (nir_is_per_vertex_io(var, stage)) {
+ if (nir_is_per_vertex_io(var, stage) || var->data.per_view) {
assert(glsl_type_is_array(type));
type = glsl_get_array_element(type);
}
unsigned var_size;
if (var->data.compact) {
+ /* If we are inside a partial compact,
+ * don't allow another compact to be in this slot
+ * if it starts at component 0.
+ */
+ if (last_partial && var->data.location_frac == 0) {
+ location++;
+ }
+
/* compact variables must be arrays of scalars */
assert(glsl_type_is_array(type));
assert(glsl_type_is_scalar(glsl_get_array_element(type)));
if (last_partial)
location++;
+ exec_list_append(&shader->variables, &io_vars);
*size = location;
}
+static uint64_t
+get_linked_variable_location(unsigned location, bool patch)
+{
+ if (!patch)
+ return location;
+
+ /* Reserve locations 0...3 for special patch variables
+ * like tess factors and bounding boxes, and the generic patch
+ * variables will come after them.
+ */
+ if (location >= VARYING_SLOT_PATCH0)
+ return location - VARYING_SLOT_PATCH0 + 4;
+ else if (location >= VARYING_SLOT_TESS_LEVEL_OUTER &&
+ location <= VARYING_SLOT_BOUNDING_BOX1)
+ return location - VARYING_SLOT_TESS_LEVEL_OUTER;
+ else
+ unreachable("Unsupported variable in get_linked_variable_location.");
+}
+
+static uint64_t
+get_linked_variable_io_mask(nir_variable *variable, gl_shader_stage stage)
+{
+ const struct glsl_type *type = variable->type;
+
+ if (nir_is_per_vertex_io(variable, stage)) {
+ assert(glsl_type_is_array(type));
+ type = glsl_get_array_element(type);
+ }
+
+ unsigned slots = glsl_count_attribute_slots(type, false);
+ if (variable->data.compact) {
+ unsigned component_count = variable->data.location_frac + glsl_get_length(type);
+ slots = DIV_ROUND_UP(component_count, 4);
+ }
+
+ uint64_t mask = u_bit_consecutive64(0, slots);
+ return mask;
+}
+
+nir_linked_io_var_info
+nir_assign_linked_io_var_locations(nir_shader *producer, nir_shader *consumer)
+{
+ assert(producer);
+ assert(consumer);
+
+ uint64_t producer_output_mask = 0;
+ uint64_t producer_patch_output_mask = 0;
+
+ nir_foreach_shader_out_variable(variable, producer) {
+ uint64_t mask = get_linked_variable_io_mask(variable, producer->info.stage);
+ uint64_t loc = get_linked_variable_location(variable->data.location, variable->data.patch);
+
+ if (variable->data.patch)
+ producer_patch_output_mask |= mask << loc;
+ else
+ producer_output_mask |= mask << loc;
+ }
+
+ uint64_t consumer_input_mask = 0;
+ uint64_t consumer_patch_input_mask = 0;
+
+ nir_foreach_shader_in_variable(variable, consumer) {
+ uint64_t mask = get_linked_variable_io_mask(variable, consumer->info.stage);
+ uint64_t loc = get_linked_variable_location(variable->data.location, variable->data.patch);
+
+ if (variable->data.patch)
+ consumer_patch_input_mask |= mask << loc;
+ else
+ consumer_input_mask |= mask << loc;
+ }
+
+ uint64_t io_mask = producer_output_mask | consumer_input_mask;
+ uint64_t patch_io_mask = producer_patch_output_mask | consumer_patch_input_mask;
+
+ nir_foreach_shader_out_variable(variable, producer) {
+ uint64_t loc = get_linked_variable_location(variable->data.location, variable->data.patch);
+
+ if (variable->data.patch)
+ variable->data.driver_location = util_bitcount64(patch_io_mask & u_bit_consecutive64(0, loc)) * 4;
+ else
+ variable->data.driver_location = util_bitcount64(io_mask & u_bit_consecutive64(0, loc)) * 4;
+ }
+
+ nir_foreach_shader_in_variable(variable, consumer) {
+ uint64_t loc = get_linked_variable_location(variable->data.location, variable->data.patch);
+
+ if (variable->data.patch)
+ variable->data.driver_location = util_bitcount64(patch_io_mask & u_bit_consecutive64(0, loc)) * 4;
+ else
+ variable->data.driver_location = util_bitcount64(io_mask & u_bit_consecutive64(0, loc)) * 4;
+ }
+
+ nir_linked_io_var_info result = {
+ .num_linked_io_vars = util_bitcount64(io_mask),
+ .num_linked_patch_io_vars = util_bitcount64(patch_io_mask),
+ };
+
+ return result;
+}